College of Science & Engineering
This group's current project will support fundamental research that dramatically advances understanding of the process-structure-properties relation in metal additive manufacturing (AM). This will be achieved by leveraging knowledge-augmented artificial intelligence (AI) to discover hidden defect mechanisms during AM processes. It is important because defects are the prevalent failure driver in nearly all AM fabricated products and dictate the part quality, but the correlation between defects and process parameters remains poorly understood due to the high dimensional operating conditions and the lack of effective predictive tools with considering process history-dependent information. To fill this gap, these researchers are testing a novel data assimilation approach, process-aware neural operators (PANO), which can merge both data and physics-based models for AI learning process. They hypothesize that this PANO framework allows for effectively utilizing various data, ranging from discrete sensor measurements to spatiotemporal information given in continuous function forms, leading to successful real-time defect prediction.
Another ongoing project is to develop a physics-informed surrogate model to simulate ice sheet melting process in large time scales by using a neural network operator learning approach, DeepOnet, together with physics-informed neural networks (PINN). The DeepOnet will be trained on high-fidelity simulation data provided by shallow ice approximation and shallow shaft approximation, and then plugged into PINN as an internal solver to accelerate long-term simulation. This coupled framework will provide 100X the speed of conventional numerical solvers.